qMTNet: Accelerated quantitative magnetization transfer imaging with artificial neural networks

Abstract

To develop a set of artificial neural networks, collectively termed qMT-Net, to accelerate data acquisition and fitting for quantitative magnetization transfer (qMT) imaging. Methods: Conventional and interslice qMT data were acquired with two flip angles at six offset frequencies from seven subjects for developing the networks and from four young and four older subjects for testing the generalizability. Two subnetworks, qMTNet-acq and qMTNet-fit, were developed and trained to accelerate data acquisition and fitting, respectively. qMTNet-2 is the sequential application of qMTNetacq and qMTNet-fit to produce qMT parameters (exchange rate, pool fraction) from undersampled qMT data (two offset frequencies rather than six). qMTNet-1 is one single integrated network having the same functionality as qMTNet-2. qMTNet-fit was compared with a Gaussian kernel-based fitting. qMT parameters generated by the networks were compared with those from ground truth fitted with a dictionarydriven approach. Results: The proposed networks achieved high peak signal-to-noise ratio (>30) and structural similarity index (>97) in reference to the ground truth. qMTNet-fit produced qMT parameters in concordance with the ground truth with better performance than the Gaussian kernel-based fitting. qMTNet-2 and qMTNet-1 could accelerate data acquisition at threefold and accelerate fitting at 5800-and 4218-fold, respectively. qMTNet-1 showed slightly better performance than qMTNet-2, whereas qMTNet-2 was more flexible for applications. Conclusion: The proposed single (qMTNet-1) and two joint neural networks (qMT-Net-2) can accelerate qMT workflow for both data acquisition and fitting significantly. qMTNet has the potential to accelerate qMT imaging for clinical applications, which warrants further investigation.

Deep Learning and Its Application to Function Approximation for MR in Medicine: An Overview

“…DNNs were used for the purpose of estimating quantitative magnetization transfer (qMT) parameters from a small number of MT images. 127 This method 127 used DNNs as a function The DNNs using images with 34 spokes per frame only generated images whose intensities were partially higher than (a), as shown in (b) and (c). This problem was not caused by adding images whose resolutions were temporally lower but spatially higher than images with 34 spokes per frame.…”

Improving quantitative MRI using self‐supervised deep learning with model reinforcement: Demonstration for rapid T1 mapping

“…Since each acquisition can take several minutes, the total scan time can be impractical for clinical settings. There has been some recent progress using deep learning-based methods to accelerate qMRI, [14][15][16][17][18][19][20][21][22][23] including MANTIS by Liu et al, 18,19 DeepDTI by Tian et al, 20 MoDL-QSM by Feng et al, 21 and DOPAMINE by Jun et al, 22 all of which use supervised learning to enable rapid MR parameter mapping from undersampled k-space data.…”

qMTNet+, an Improved qMTNet with Residual Connection for Accelerated Quantitative Magnetization Transfer Imaging

“…With an increasing popularity of deep learning and neural networks, a family of artificial neural networks (ANN) called qMTNet have been recently proposed to accelerate both data acquisition and fitting time for qMT imaging (20). Building blocks of qMTNet consist of qMTNet-acq, a convolutional neural network to produce unsampled MT images from acquired ones, and qMTNet-fit, a multilayer perceptron (MLP) to fit qMT parameters from fully-sampled MT data.…”

“…Motivated by (20), in this study, we proposed qMTNet + , an improved version of qMTNet. Unlike qMTNet that separates the acceleration process into two subnetworks, qMTNet + is a single network that is trained end-to-end to produce both qMT parameters and un-acquired MT offset images from under-sampled data.…”